I have recently encountered a recurring issue in Google AI Studio where the coding agent reports that requested changes have been completed successfully, but the changes are not actually reflected in the Preview pane or the live production application.
This has happened multiple times recently.
A typical workflow is:
-
I ask the AI Studio coding agent to modify an existing application.
-
The agent reports that the implementation is complete.
-
It may run local tests and report that they pass.
-
I publish the application from Google AI Studio to the live production environment.
-
However, the Preview and/or live production application still behaves exactly as before.
In one recent case, the agent added a new API filter and successfully tested it in its local environment. The local test returned the expected filtered results and validation behaviour.
However, after publishing from AI Studio, testing the actual live production endpoint showed that:
-
the new response field was still missing,
-
the new filter was not being applied,
-
and the new validation behaviour was not present.
The live application was effectively still running the previous implementation.
What makes this especially confusing is that AI Studio continues to report the task as successfully completed, even though the published application does not reflect the changes.
I am intentionally leaving out project-specific details because this is a public forum, but the issue appears to involve a mismatch between:
-
the AI agent’s workspace,
-
the Preview environment,
-
the saved project state,
-
the build output,
-
and the version actually published to production.
The main concern is that the coding agent does not clearly distinguish between:
-
code changed locally,
-
code saved successfully,
-
build completed,
-
Preview updated,
-
deployment/publishing completed,
-
and the live production version actually running the new code.
For example, reporting “implemented and tested successfully” can give the impression that the task is finished, while the live application remains unchanged even after publishing.
It would be very helpful if AI Studio could verify the deployed version after publishing and explicitly report the deployment state, for example:
-
Implemented locally
-
Saved
-
Build successful
-
Preview updated
-
Published successfully
-
Live production verified
If the live application is still serving the previous version, the agent should not report the task as fully completed.
This seems to be a recent reliability issue for me, as the same workflow was previously much more consistent.
Has anyone else recently experienced AI Studio reporting successful code changes or publishing, while Preview or the live production application continues to show the previous version?